Using Deep Learning for Image-Based Plant Disease Detection

Resources:

• Dataset

• Additional dataset will be relesed

• Models

Objective

• Train and Evaluate different DNN Models for plant disease detection problem

• To tackle the problem of scarce real-life representative data, experiment with different generative networks and generate more plant leaf image data

• Implement segmentation pipeline to avoid misclassification due to unwanted input

Approches for Solving the papers realtime Detection Problem

phase 1 : implement the paper

phase 2 : do analysis on the paper and identify the type of data problem

phase 3 : experiment and if possible generate appropriate data using the data to train the model again

Plant_Disease_Detection_Benchmark_models

• Train and test different prediction models to get a baseline accuracy to compare to and see progress

Plant_Disease_Detection_gan_experiments

• experiment with different generative networks to see their generative capability and if the output can be used to train more robust models

leaf-image-segmentation-segnet

• segmentation pipeline using VGGSegNet Architecture

leaf-image-segmentation

• histogram based segmentation Pipline

python main.py IMAGE_FILE [--segment] [--species SPECIES_TYPE] [--model PREDICTION_MODEL]

Arguments:
	IMAGE_FILE    Path of the image file
	--segment     If specified perform segmentation on the image before prediction
	--species     If the plant species on the image is priorly known. One of the following species:  Apple, Cherry, Corn, Grape, Peach, Pepper, Potato, Strawberry, Sugercane, Tomato
	--model       What models do you want to use, vgg or inceptionv3

# you can remove a part of arguments except image path

>>  python main.py 'test/a.jpg' --segment --species 'apple' --model 'inceptionv3'

• Before using that make sure you download the weights from here for Inception_V3 and here for VGG Models and extract all and put it in Plant_Disease_Detection_Benchmark_models/Models/ folder.

• This will segment the image and predict the output class based on that. Segmented image will be saved as the file name with "_marked" suffix before the file extension.

• The images are trained with segmented network and lower performance on unsegmented dataset is expected.

• You can check the segmentation accuracy from saved image.

• Fill this form for bulk model access grants and future update notification.